1: \documentclass{emulateapj}

2: %\documentclass[12pt, preprint]{aastex}

3: 

4: \usepackage{float}

5: \usepackage{amsmath}

6: \usepackage{epsfig,floatflt}

7: \usepackage{subfigure}

8: 

9: \begin{document}

10: 

11: \title{Bayesian analysis of the low-resolution polarized 3-year WMAP

12:   sky maps}

13: 

14: \author{H.\ K.\ Eriksen\altaffilmark{1,2,3}, Greg

15:   Huey\altaffilmark{4,5,6}, A. J. Banday\altaffilmark{7}, K. M.

16:   G\'{o}rski\altaffilmark{5,6,8}, J. B.  Jewell\altaffilmark{4,5}, \\I.\

17:   J.\ O'Dwyer\altaffilmark{4,5}, B.\ D.\ Wandelt\altaffilmark{4,7,9}}

18: 

19: \altaffiltext{1}{email: h.k.k.eriksen@astro.uio.no}

20: 

21: \altaffiltext{2}{Institute of Theoretical Astrophysics, University of

22: Oslo, P.O.\ Box 1029 Blindern, N-0315 Oslo, Norway}

23: 

24: \altaffiltext{3}{Centre of

25: Mathematics for Applications, University of Oslo, P.O.\ Box 1053

26: Blindern, N-0316 Oslo}

27: 

28: \altaffiltext{4}{Department of Physics, University of Illinois,

29:   Urbana, IL 61801}

30: 

31: \altaffiltext{5}{Jet Propulsion Laboratory, 4800 Oak

32:   Grove Drive, Pasadena CA 91109} 

33: 

34: \altaffiltext{6}{California Institute of Technology, Pasadena, CA

35:   91125} 

36: 

37: \altaffiltext{7}{Max-Planck-Institut f\"ur Astrophysik,

38: Karl-Schwarzschild-Str.\ 1, Postfach 1317, D-85741 Garching bei

39: M\"unchen, Germany}

40: 

41: \altaffiltext{8}{Warsaw University Observatory, Aleje Ujazdowskie 4, 00-478 Warszawa,

42:   Poland}

43: 

44: \altaffiltext{9}{Astronomy Department, University of Illinois at

45:   Urbana-Champaign, IL 61801-3080}

46: 

47: \date{Received - / Accepted -}

48: 

49: \begin{abstract}

50: We apply a previously developed Gibbs sampling framework to the

51: foreground corrected 3-yr WMAP polarization data and compute the power

52: spectrum and residual foreground template amplitude posterior

53: distributions. We first analyze the co-added Q- and V-band data, and

54: compare our results to the likelihood code published by the WMAP

55: team. We find good agreement, and thus verify the numerics and data

56: processing steps of both approaches. However, we also analyze the Q-

57: and V-band separately, allowing for non-zero EB cross-correlations and

58: including two individual foreground template amplitudes tracing

59: synchrotron and dust emission. In these analyses, we find tentative

60: evidence of systematics: The foreground tracers correlate with each of

61: the Q- and V-band sky maps individually, although not with the

62: co-added QV map; there is a noticeable negative EB cross-correlation

63: at $\ell \lesssim 16$ in the V-band map; and finally, when relaxing

64: the constraints on EB and BB, noticeable differences are observed

65: between the marginalized band powers in the Q- and V-bands. Further

66: studies of these features are imperative, given the importance of the

67: low-$\ell$ EE spectrum on the optical depth of reionization $\tau$ and

68: the spectral index of scalar perturbations $n_{s}$.

69: \end{abstract}

70: 

71: \keywords{cosmic microwave background --- cosmology: observations --- 

72: methods: numerical}

73: 

74: \maketitle

75: 

76: \section{Introduction}

77: 

78: One of the most remarkable results in the 3-yr data release from the

79: Wilkinson Microwave Anisotropy Probe (WMAP) experiment

80: \citep{hinshaw:2007, page:2007} was the detection of large-scale

81: E-mode polarization at millimeter wavelengths. This was interpreted as

82: the theoretically predicted signature of reionization, and allowed the

83: WMAP team to set new and tighter constraints on the optical depth of

84: reionization $\tau$. In turn, the well-known degeneracy between $\tau$

85: and the spectral index of primordial scalar perturbations $n_{s}$ was

86: broken. The final outcome was a claimed detection of $n_{s} \ne 1$ at

87: a statistical significance of almost $3\sigma$ \citep{spergel:2007}.

88: 

89: One should bear in mind, however, the great potential for systematics

90: effects in both the temperature and polarization measurements. For

91: instance, the precise level of power contribution from unresolved

92: point sources affects $n_{s}$ directly. An independent analysis of

93: this particular issue by \citet{huffenberger:2006} showed that the

94: initial point source amplitude quoted by the WMAP team was indeed too

95: high, which biased $n_{s}$ to low values. Similarly, on large scales

96: the likelihood approximation used by the WMAP team was biased high

97: \citep{eriksen:2007}, which also biased $n_s$ low. After these

98: corrections, the statistical significance of $n_s \ne 1$ dropped to

99: $\sim2\sigma$.

100: 

101: For polarization the situation may be even more serious due to the

102: strong sensitivity of $\tau$ and $n_s$ on the low-$\ell$ EE spectrum,

103: combined with the low signal-to-noise ratio of the WMAP

104: data. Systematic effects, both from the instrument itself

105: \citep{jarosik:2007} and from non-cosmological foregrounds

106: \citep{kogut:2007}, are much more likely to affect the results, and we

107: are also much less likely to detect them. It is therefore imperative

108: to carefully check both the data and the analysis methods, in order to

109: build up confidence in the final cosmological results. In this Letter,

110: we start this task by computing the low-$\ell$ EE, EB, BB and

111: foreground template amplitude posterior distributions from the WMAP

112: data.

113: 

114: \section{Method}

115: \label{sec:method}

116: 

117: We use a previously introduced Gibbs sampling framework (Jewell et

118: al.\ 2004, Wandelt et al.\ 2004, Eriksen et al.\ 2004, Larson et al.\

119: 2007; hereafter JWEL) to estimate the posterior distributions. For

120: full details on the method, we refer the interested reader to the

121: quoted papers, and only summarize the principles here.

122: 

123: \begin{figure*}[t]

124: 

125: \mbox{\epsfig{figure=f1.eps,width=\linewidth,clip=}}

126: 

127: \caption{Comparison of single-$\ell$ EE marginal posteriors from the

128:   WMAP code (black distributions) and the Gibbs sampler (red

129:   distributions) using the QV map. The EB and BB power spectra are

130:   held at zero.}

131: \label{fig:EE_comparison}

132: \end{figure*}

133: 

134: First we define our notation. The desired distribution is denoted

135: $P(\mathbf{s}, C_{\ell}, \mathbf{f}|\mathbf{d})$, where $\mathbf{s}$

136: is the CMB signal, $C_{\ell} = \{C_{\ell}^{\textrm{EE}},

137: C_{\ell}^{\textrm{EB}}, C_{\ell}^{\textrm{BB}}\}$ is the CMB power

138: spectrum, $\mathbf{f}$ is a set of foreground template amplitudes, and

139: $\mathbf{d}$ are the data.

140: 

141: The Gibbs sampler is a Markov Chain Monte Carlo method, and, as such,

142: maps out the full posterior by drawing samples from it. While direct

143: evaluation or sampling from the posterior $P(C_{\ell }| \mathbf{d})$

144: requires inversion of a prohibitively large matrix, the Gibbs sampling

145: scheme \citep{gelfand:1990} uses the conditional densities of the

146: joint posterior $P(C_{\ell}, \mathbf{s}|\mathbf{d})$ which is

147: computationally feasible to sample from. The algorithm may thus be

148: described by the following sampling steps,

149: \begin{eqnarray}

150:   \mathbf{s}^{i+1} & \leftarrow & P(\mathbf{s}|C_{\ell}^{i},

151:   \mathbf{f}^{i}, \mathbf{d}) \\

152:   C_{\ell}^{i+1} & \leftarrow & P(C_{\ell}|\mathbf{s}^{i+1},

153:   \mathbf{f}^{i}, \mathbf{d}) \\

154:   \mathbf{f}^{i+1} & \leftarrow & P(\mathbf{f}|\mathbf{s}^{i+1},

155:   C_{\ell}^{i+1}, \mathbf{d}).

156: \end{eqnarray}

157: Here the symbol $\leftarrow$ indicates sampling from the conditional

158: distribution on the right hand side, which can be accomplished without

159: inverting the signal-plus-noise covariance matrix (see JWEL for

160: details). For the foreground template amplitude distribution, we note

161: that the required algorithm is identical to that employed for sampling

162: monopole and dipole amplitudes \citep{eriksen:2004}.

163: 

164: \section{Data}

165: \label{sec:data}

166: 

167: We consider only the low-resolution foreground-corrected 3-yr WMAP

168: polarization data in this Letter, as provided on

169: LAMBDA\footnote{http://lambda.gsfc.nasa.gov}. These come in the form

170: of three HEALPix\footnote{http://healpix.jpl.nasa.gov} sky maps,

171: pixelized at $N_{\textrm{side}} = 16$, each having 3072 pixels in both

172: Stoke's Q and U. The WMAP P06 sky cut is imposed on the data, leaving

173: only 2267 pixels for the analysis. Two frequency bands are included,

174: namely Q-band (41 GHz) and V-band (61 GHz). In addition, we analyze

175: the co-added map (denoted QV), and also the two frequency

176: maps jointly but not co-added (denoted Q+V). All maps are provided

177: with a full noise covariance matrix \citep{jarosik:2007},

178: appropriately corrected for the P06 sky cut and removal of foreground

179: templates. The units used in this paper are thermodynamic

180: $\mu\textrm{K}$.

181: 

182: For foreground marginalization, we adopt two individual

183: templates. First, we use the K--Ka difference map, smoothed to

184: $10^{\circ}$ FWHM resolution to reduce noise contributions, as a

185: tracer of synchrotron emission. Second, for dust emission we adopt the

186: low-noise template developed by the WMAP team for their foreground

187: correction procedure \citep{page:2007}. Note that the specific shape

188: of these templates are of minor importance; if the provided sky maps

189: are free of foregrounds, they should not correlate significantly with

190: any non-CMB map.

191: 

192: We compare our results to the official WMAP likelihood

193: code\footnote{version v2p2p2}, also available from LAMBDA. To map out

194: the appropriate posteriors, we have written a special-purpose MCMC

195: wrapper around this likelihood code.

196: 

197: \section{Results}

198: \label{sec:results}

199: 

200: \subsection{Numerical verification of the WMAP likelihood}

201: \label{sec:verification}

202: 

203: 

204: The first case considered is that adopted by the WMAP likelihood code,

205: namely the co-added QV map. For this analysis, we fix the EB and BB

206: spectra to zero, and map out the corresponding marginalized EE

207: posteriors $\ell$-by-$\ell$, both with the Gibbs sampler and by the

208: WMAP-based MCMC code.

209: 

210: \begin{figure}[tb]

211: 

212: \mbox{\epsfig{figure=f2.eps,width=\linewidth,clip=}}

213: 

214: \caption{Marginal posterior distributions for the EE bin power between

215: $\ell=2$--6.} 

216: \label{fig:EE_bin_power}

217: \end{figure}

218: 

219: The results from this exercise are shown in Figure

220: \ref{fig:EE_comparison}. The agreement between the two approaches is

221: very good, and this is an important validation of the WMAP data

222: processing method: First, we analyze the data at their native

223: $N_{\textrm{side}} = 16$ resolution, while the WMAP team analyze maps

224: downgraded to $N_{\textrm{side}} = 8$. Second, they marginalize over a

225: single total foreground template, while we marginalize over the K--Ka

226: difference map and a dust template. Third, we use a Gibbs sampler for

227: the numerical work, while the WMAP team uses a brute-force likelihood

228: evaluator. None of these differences affects the low-$\ell$ EE

229: spectrum peak visibly, as will be quantified more precisely in the

230: next section.

231: 

232: \subsection{Generalized analysis}

233: \label{sec:stability}

234: 

235: \begin{deluxetable}{lcccc}

236: \tablewidth{0pt} 

237: \tabletypesize{\small} 

238: \tablecaption{Marginalized EE and BB band powers\label{tab:band_power}}

239: \tablecolumns{5}

240: \tablehead{ & \multicolumn{2}{c}{EE power ($10^{-1}\,\mu\textrm{K}^2$)} &

241:   \multicolumn{2}{c}{BB power ($10^{-1}\,\mu\textrm{K}^2$)} \\ Data set & $\ell=2$--6 & $\ell=2$--20 &$\ell=2$--6 & $\ell=2$--20 

242: }

243: 

244: \startdata

245: \cutinhead{EE free; EB = BB = 0}

246: WMAP & $1.1^{+0.9}_{-0.5}$ & $0.64^{+0.46}_{-0.34}$ & \nodata & \nodata \\

247: QV-band & $1.2^{+0.9}_{-0.6}$ & $0.67^{+0.39}_{-0.38}$ &

248: \nodata & \nodata \\

249: Q+V-band & $1.1^{+0.8}_{-0.6}$ & $0.65^{+0.38}_{-0.35}$ &

250: \nodata & \nodata \\

251: Q-band & $1.0^{+1.0}_{-0.8}$ & $0.36^{+0.67}_{-0.36}$ &

252: \nodata & \nodata \\

253: V-band & $1.3^{+1.2}_{-0.9}$ & $1.2^{+0.9}_{-0.7}$ &

254: \nodata & \nodata \\

255: 

256: \cutinhead{EE, BB free; EB = 0}

257: WMAP & $0.94^{+0.76}_{-0.58}$ & $0.63^{+0.44}_{-0.37}$ &

258: $<0.70$ & $<0.40$  \\

259: QV-band & $1.1^{+0.8}_{-0.6}$ & $0.61^{+0.38}_{-0.37}$ &

260: $<0.57$ & $<0.26$ \\

261: Q+V-band & $1.1^{+0.8}_{-0.6}$ & $0.57^{+0.40}_{-0.31}$ &

262: $<0.58$ & $<0.30$ \\

263: Q-band & $0.3^{+1.3}_{-0.3}$ & $0.23^{+0.68}_{-0.23}$ &

264: $0.3^{+1.2}_{-0.3}$ & $<0.71$ \\

265: V-band & $1.4^{+1.4}_{-0.9}$ & $1.1^{+1.0}_{-0.7}$ &

266: $<0.94$ & $<0.51$ \\

267: 

268: \cutinhead{EE, EB, BB free}

269: QV-band & $1.4^{+0.9}_{-0.7}$ & $0.65^{+0.41}_{-0.30}$ &

270: $0.30^{+0.60}_{-0.30}$ & $0.1^{+0.3}_{-0.1}$ \\

271: Q+V-band & $1.3^{+0.9}_{-0.7}$ & $0.66^{+0.43}_{-0.30}$ &

272: $0.31^{+0.59}_{-0.31}$ & $0.1^{+0.3}_{-0.1}$ \\

273: Q-band & $1.1^{+1.1}_{-0.8}$ & $0.54^{+0.65}_{-0.36}$ &

274: $0.7^{+1.3}_{-0.6}$ & $0.5^{+0.7}_{-0.4}$ \\

275: V-band & $1.8^{+1.6}_{-1.1}$ & $1.5^{+0.9}_{-0.9}$ &

276: $0.47^{+0.93}_{-0.47}$ & $0.4^{+0.6}_{-0.4}$ 

277: 

278: \enddata

279: 

280: \tablecomments{Values indicate either the posterior mode and upper

281: and lower 68\% confidence interval, or the upper 68\% confidence

282: limits. If the lower error bar equals the posterior mode value, a

283: non-zero peak is detected, but at a significance of less than

284: 68\%.}

285: 

286: \end{deluxetable}

287: 

288: Having validated the numerics and data processing steps in both

289: procedures, we now expand the analysis and allow for non-zero EB

290: and/or BB spectra. We also compute the posteriors for each frequency

291: band separately and jointly. For presentational reasons, we report

292: only band powers in EE and BB between $\ell=2$--6 and $\ell=2$--20,

293: respectively. Cases of special interest are treated separately in

294: subsequent sections. 

295: 

296: In order to achieve good convergence, $10^6$ samples were generated

297: for the non-zero EB, $\ell=2$--20 cases. For all other cases, $10^5$

298: samples were generated. The CPU time to generate one sample was $\sim2$

299: seconds.

300: 

301: The results from these computations are summarized in Table

302: \ref{tab:band_power}. The EE posteriors with fixed EB = BB = 0 are

303: shown in Figure \ref{fig:EE_bin_power}. Again, note the excellent

304: agreement between the WMAP results and the QV and Q+V cases in the two

305: top sections.

306: 

307: However, even though the joint QV analyses agree well, the picture is

308: considerably less clear when it comes to single bands and relaxed EB

309: constraints. On the one hand, there appears to be more EE power in the

310: V-band data than in the Q-band data. On the other hand, there may seem

311: to be small traces of BB power in the Q-band data.

312: 

313: The BB posteriors develop a peak away from $C^{\textrm{BB}}_{\ell}=0$

314: in the EB$\ne0$ case. This is not surprising. Since the signal

315: covariance matrix is positive definite, one must have

316: $C_{\ell}^{\textrm{EE}} C_{\ell}^{\textrm{BB}} >

317: (C_{\ell}^{\textrm{EB}})^2$. Therefore, when marginalizing over

318: $C_{\ell}^{\textrm{EB}}$, a non-zero peak emerges in both the EE and

319: BB spectra individually.

320: 

321: \subsubsection{Foreground amplitude posteriors}

322: 

323: \begin{figure*}[t]

324: 

325: \mbox{\epsfig{figure=f3.eps,width=\linewidth,clip=}}

326: 

327: \caption{The foreground template amplitude marginal posteriors for the

328:   Q-, V- and QV-bands. For this plot, the EB and BB power spectra were

329:   set to zero.}

330: \label{fig:foregrounds}

331: \end{figure*}

332: 

333: %Q-band, K-Ka: 4.4\%

334: %Q-band, dust: 93.0\%

335: 

336: %V-band, K-Ka: 74.1\%

337: %V-band, dust: 1.5\%

338: 

339: %QV-band, K-Ka: 21.4\%

340: %QV-band, dust: 33.4\%

341: 

342: In Figure \ref{fig:foregrounds} we show the foreground template

343: amplitude posteriors for the Q-, V- and QV-band data, for the case

344: with fixed EB = BB = 0. Although no signs of significant residual

345: foregrounds are observed in the co-added QV band, apparently

346: confirming the fits made by the WMAP team, the same is not true for

347: each band individually. On the contrary, non-zero correlations are

348: seen in both the Q- and V-band data individually. 

349: 

350: For the Q-band data, the marginal best-fit K--Ka amplitude is

351: $A_{\textrm{s}} = -0.027^{+0.014}_{-0.17}$, different from zero at

352: $2\sigma$. The best-fit dust amplitude is $A_{\textrm{d}} =

353: 15.7^{+8.7}_{-10.7}$. For the V-band data, the best-fit dust amplitude

354: is $A_{\textrm{d}} = -24.1^{+10.3}_{-11.3}$, 2.3$\sigma$ away from

355: zero. The K--Ka amplitude is $A_{\textrm{s}} =

356: 0.011^{+0.015}_{-0.018}$.

357: 

358: These results may be compared to Table 4 of \citet{page:2007}. The

359: main difference between the two analyses is that while we apply the

360: conservative P06 mask to the data, \citet{page:2007} apply the much

361: more aggressive processing mask described by \citet{jarosik:2007}. The

362: two analyses remove 26.4\% and 5.7\% of the sky,

363: respectively. Considering that all cosmological analyses are carried

364: out with the P06 mask, and that variations in the synchrotron spectral

365: index are observed between the galactic plane and the high latitudes

366: \citep{kogut:2007}, it is not immediately clear to us why the more

367: aggressive mask was chosen for this task by the WMAP team. The

368: improvement in raw $\chi^2$ after further correcting the ``cleaned''

369: WMAP maps for these residuals is -5.4 for the Q-band and -3.3 for the

370: V-band, respectively.

371: 

372: Two other, minor differences are that we take into account both the

373: full noise and CMB covariance matrices, while the WMAP team includes

374: only a white noise covariance approximation. Note that this carries no

375: extra cost within the Gibbs sampling framework.

376: 

377: \subsubsection{ E$\times$B cross-correlation spectra}

378: 

379: \begin{figure}[t]

380: 

381: \mbox{\epsfig{figure=f4.eps,width=\linewidth,clip=}}

382: 

383: \caption{The E$\times$B cross-power spectrum. The gray region shows

384:   the $1\sigma$ confidence region around the V-band spectrum (blue

385:   curve); note that the marginal EB posterior is strongly non-Gaussian

386:   with heavy tails. The green data points show the WMAP EB spectrum

387:   computed with a pseudo-$C_{\ell}$ estimator with Gaussian error

388:   bars.}

389: \label{fig:EB_correlation}

390: \end{figure}

391: 

392: %Another powerful probe of systematics is the E$\times$B

393: %cross-spectrum, which is expected to be zero for all common

394: %cosmological modes. The existence of such correlations in the data

395: %would therefore strongly imply the presence of uncorrected

396: %systematics. 

397: 

398: %A second case of particular interest is a non-zero EB correlation in

399: %the V-band sky map. 

400: We find evidence for a non-zero EB correlation in the V-band sky map.

401: This spectrum is shown in Figure

402: \ref{fig:EB_correlation} for the Q- and V-band sky maps individually,

403: together with the pseudo-$C_{\ell}$ EB QV-band spectrum computed by

404: the WMAP team. Note the consistently negative correlation observed in

405: the V-band. Although the significance of the negative correlation is

406: not more than a few tenths per multipole, and the joint significance

407: is not more than 1 to 2$\sigma$ depending on binning scheme, it is

408: observed consistently in every multipole up to $\ell=17$. A similar,

409: though weaker, trend is seen in the pseudo-spectrum computed by the

410: WMAP team from the QV combination. 

411: 

412: 

413: \section{Conclusions}

414: \label{sec:conclusions}

415: 

416: We have performed a Bayesian analysis of the low-resolution

417: foreground-corrected 3-yr WMAP polarization data using a previously

418: described methodology based on the Gibbs sampler. By doing so, we

419: validated the numerical implementation of the official WMAP likelihood

420: code, as well as their procedure for degradation of map

421: resolution. However, when generalizing the analysis to allow for

422: non-zero EB and BB power spectrum components, and also considering

423: individual frequency bands, we found several issues that may be

424: important for the cosmological interpretation of these data.

425: 

426: First and foremost, when relaxing the constraints on EB and BB,

427: noticeable differences between the Q- and V-band posteriors are

428: observed. Specifically, we find generally more EE power in the V-band

429: data than in the Q-band data, but also perhaps some hints of BB power

430: in the Q-band data. At the same time, we have also found a negative EB

431: correlation in the V-band map, as well as residual foregrounds in both

432: maps.

433: 

434: If these tentative findings are confirmed by future experiments or

435: additional years of WMAP observations, significant shifts in

436: cosmological parameters could be the result. For example, if the

437: V-band data alone were used for the WMAP3 analysis instead of the QV

438: combination, the amplitude of the $\ell=2$--6 EE detection would increase

439: by 20--50\%, depending on whether BB is allowed to vary or

440: not. Consequently, $\tau$ could increase from 0.09 to $\sim0.12$, and

441: $n_{\textrm{s}}$ by a percent or two, comparable to its current

442: nominal statistical uncertainty of 0.015, from $\sim0.97$ to

443: $\sim0.98$.

444: 

445: As discussed in the introduction, systematics are a serious concern

446: for both the temperature and polarization data for both $\tau$ and

447: $n_s$. It is important to bear in mind that the currently quoted

448: uncertainties on these quantities often found in the literature are

449: statistical only. The unknown systematic uncertainties may turn out to

450: be non-negligible for the currently available data sets, and, in

451: particular, we believe it is too early to draw any firm conclusions

452: concerning the precise value of $n_s$. Fortunately, Planck will

453: clarify these issues in the near future.

454: 

455: \begin{acknowledgements}

456:   We wish to thank David Larson for his contributions during the early

457:   phases of this project. We acknowledge use of the Legacy Archive for

458:   Microwave Background Data Analysis (LAMBDA). We acknowledge use of

459:   the HEALPix software \citep{gorski:2005} and analysis package for

460:   deriving the results in this paper. This work was partially

461:   performed at the Jet Propulsion Laboratory, California Institute of

462:   Technology, under a contract with the National Aeronautics and Space

463:   Administration. HKE acknowledges financial support from the Research

464:   Council of Norway. BDW acknowledges support by NSF grant

465:   no. 0507676, the Friedrich Wilhelm Bessel Research Prize by the

466:   Alexander v. Humboldt foundation and NASA subcontract no. 1236748.

467: \end{acknowledgements}

468: 

469: 

470: 

471: \begin{thebibliography}{}

472: 

473: \bibitem[Eriksen et al.(2004)]{eriksen:2004} 

474: Eriksen, H.~K., et al.\ 2004, \apjs, 155, 227

475: 

476: \bibitem[Eriksen et al.(2007)]{eriksen:2007} 

477: Eriksen, H.~K., et al.\ 2007, \apj, 641, 665

478: 

479: \bibitem[Gelfand \& Smith(1990)]{gelfand:1990}

480: Gelfand, A. E., \& Smith, A. F. M. 1990, J. Am. Stat. Asso., 85, 398

481: 

482: \bibitem[G{\'o}rski et al.(2005)]{gorski:2005} 

483:   G{\' o}rski, K.~M., Hivon, E., Banday, A.~J., Wandelt, B.~D.,

484:   Hansen, F.\,K., Reinecke, M., \& Bartelmann, M. 2005, \apj, 622, 759

485: 

486: \bibitem[Hinshaw et al.(2007)]{hinshaw:2007} Hinshaw, G., et al.\ 

487: 2007, ApJ, in press [astro-ph/0603451]

488: 

489: \bibitem[Huffenberger et al.(2006)]{huffenberger:2006} Huffenberger, 

490: K.~M., Eriksen, H.~K., \& Hansen, F.~K.\ 2006, \apjl, 651, L81 

491: 

492: \bibitem[Jarosik et al.(2007)]{jarosik:2007} Jarosik, N., et al.\ 

493: 2007, ApJ, in press, [astro-ph/0603452]

494: 

495: \bibitem[Jewell et al.(2004)]{jewell:2004} 

496:   Jewell, J., Levin, S., \& Anderson, C.  H.  2004, \apj, 609, 1

497: 

498: \bibitem[Kogut et al.(2007)]{kogut:2007} Kogut, A., et al.\ 2007, 

499: ApJ, submitted [arXiv:0704.3991] 

500: 

501: \bibitem[Larson et al.(2007)]{larson:2007} Larson, D.~L., Eriksen, 

502: H.~K., Wandelt, B.~D., G{\'o}rski, K.~M., Huey, G., Jewell, J.~B., \& 

503: O'Dwyer, I.~J.\ 2007, \apj, 656, 653 

504: 

505: \bibitem[Page et al.(2007)]{page:2007} Page, L., et al.\ 2007, 

506: ApJ, in press, [astro-ph/0603450]

507: 

508: \bibitem[Spergel et al.(2007)]{spergel:2007} Spergel, D.~N., et al.\ 

509: 2007, ApJ, in press, [astro-ph/0603449]

510:  

511: \bibitem[Wandelt et al.(2004)]{wandelt:2004} 

512:   Wandelt, B.~D., Larson, D.~L., \& Lakshminarayanan, A.\ 2004, \prd,

513:   70, 083511

514: 

515: 

516: \end{thebibliography}

517: 

518: 

519: 

520: %\begin{deluxetable*}{lcccc}

521: %\tablewidth{0pt} 

522: %\tabletypesize{\small} 

523: %\tablecaption{Low-$\ell$ EE band power in the WMAP data\label{tab:band_power}}

524: %\tablecolumns{5}

525: %\tablehead{         & WMAP & \multicolumn{3}{c}{Gibbs sampler} \\

526: %           Data set & WMAP & Q+V-band & Q-band & V-band}

527: 

528: %\startdata

529: 

530: %\cutinhead{With foreground template marginalization}

531: %A & B & C & D & E

532: %\enddata

533: 

534: %\tablecomments{}

535: 

536: %\end{deluxetable*}

537: 

538: 

539: 

540: 

541: 

542: 

543: 

544: \end{document}